Massive MIMO: Fundamentals, Opportunities and Challenges: MM YS

MM
YS
Massive MIMO:
Fundamentals, Opportunities and Challenges
Erik G. Larsson
June 9, 2013
Div. of Communication Systems
Dept. of Electrical Engineering (ISY)
Link oping University
Link oping, Sweden
www.commsys.isy.liu.se
With thanks to my team and collaborators:
Hien Q. Ngo (LiU, Sweden)
Antonios Pitarokoilis (LiU)
Hei Victor Cheng (LiU)
Daniel Persson (LiU)
Fredrik Rusek (Lund, Sweden)
Ove Edfors (Lund)
Buon Kiong Lau (Lund)
Fredrik Tufvesson (Lund)
Thomas L. Marzetta (Bell Labs/Alcatel-Lucent, USA)
Saif Mohammed (IIT/Delhi, India)
Christoph Studer (Rice Univ., USA)
1/44
Erik G. Larsson
Massive MIMO: Fundamentals, Opportunities and Challenges
Communication Systems
Massive MIMO
M=
x100
antennas!
K
terminals
k=1
k=K
Massive multiuser MIMO (MISO):
M K 1 (think 100 10 or 500 50)
coherent, but simple, processing
Potential to dramatically improve rate & reliability
Potential to drastically scale down TX power
Not only theory, at least one known testbed (64 10)

2/44
Erik G. Larsson
Large MIMO Deployment Scenarios
Reduce bulky items (coax)
Each antenna unit simple (low accuracy)
Resilience against individual failures (hotswapping)
Potential economy of scale in manufacturing
Grid-free powering
3/44
Erik G. Larsson
Massive MIMO Operation
Not enough resources for pilots & CSI feedback, so operate in TDD.
On the uplink,
acquire CSI from uplink pilots and/or blindly from data
detect symbols
M K linear processing (MRC, ZF, MMSE) nearly optimal
On the downlink,
use CSI obtained on the uplink
make necessary adjustments based on reciprocity calibration
apply multiuser MIMO precoding
simple precoders desirable (and very good!): MRT, ZF, MMSE, ...
MRC/MRT operation
intracell interference will appear as noise
1 bps/Hz/terminal; K bps/Hz/terminal total
distributed implementation
ZF/MMSE operation
can cancel out intra cell interference
computationally more demanding

4/44
Erik G. Larsson
MRT versus ZF precoding
5/44
Erik G. Larsson
Massive MIMO Opportunities
Multiplexing gain K, Array power gain M (ideally)
Rely on the law of large numbers
average out fast fading and thermal noise
Channel hardens h
2
/M constant
no FD scheduling, full BW to all terminals, simple MAC
almost no air-interface latency
In the 1 bps/Hz/t regime, many impairments drown in thermal noise
M K unused degrees of freedom (e.g. 500 50 = 450)

6/44
Erik G. Larsson
Massive MIMO Challenges and Questions
Multiplexing gain can only materialize if channel responses are

(nearly) orthogonal
Array gain relies on coherency getting CSI is the main thing
HW power consumption (RF) must scale fast enough with P

HW power consumption (BB) must scale slow enough with M
Scaling down P thermal noise eventually limits performance.
P large enough, interference limits performance
intracell interference
intercell interference
7/44
Erik G. Larsson
Limits Imposed by Propagation
8/44
Erik G. Larsson
Favorable propagation in Point-to-Point MIMO
M K MIMO link H, M K
Favorable propagation (f.p.) if

H
H
H I
2
1
= . . . =
2
K
For H
2
= constant, log
I +
1
N
0
H
H
H
max if f.p.
max
min
_
2
= extra power needed to use all eigenmodes
H
mk
zero mean & i.i.d., and M K f.p.
9/44
Erik G. Larsson
Favorable propagation in MU-SIMO
Favorable propagation if
1
M
g
H
i
g
j
0 for i = j
g
i
2
depends on path loss and shadow fading
10/44
Erik G. Larsson
I.i.d. channels give favorable propagation
tail
no tail
~20 dB
~3 dB
11/44
Erik G. Larsson
Do we have favorable propagation in practice?
Our partners at Lund Univ., Sweden have conducted unique

measurements [RPL2013,GERT2011,GTER2012].
Indoor 128-ant. (4x16 dual-pol.) array. 3 users indoor, 3 outdoor.
2.6 GHz CF, 50 MHz BW, 100 snapshots (10m).
Normalized to retain only small-scale fading.

12/44
Erik G. Larsson
Lund measurements, example of results
tail
no tail
~26 dB
~7dB
13/44
Erik G. Larsson
Summary
Assumptions on f.p. have substantial support in measurements
I.i.d. model for small-scaled fading appears reasonable
More details and models in F. Tufvessons tutorial next

14/44
Erik G. Larsson
Limits Imposed by Noise and Intracell Interference
with Linear Processing
15/44
Erik G. Larsson
Single-Cell (Noise-Limited) Uplink
M antennas
K terminals
Power (SNR) per terminal: P
i.i.d. Rayleigh fading
Coherence interval: T symbols (e.g. 2 100kHz 1ms = 200)
K mutually orthogonal pilots of length (K T)

pilots data
t
T-t
MMSE channel estimation

16/44
Erik G. Larsson
Spectral-energy eciency tradeo
Sum-spectral eciency bounds [NLM2013]:

R =
_
_
_
_
1

T
_
K log
2
_
1 +
(M1)P
2
(K1)P
2
+(+K)P+1
_
, for MRC
_
1

T
_
K log
2
_
1 +
(MK)P
2
(+K)P+1
_
, for ZF
Energy eciency:
R
P
SISO system: K = 1, M = 1
max
P,
, s.t. R = const.
17/44
Erik G. Larsson
UL, T = 200, SISO reference, M = K = 1
0 10 20 30 40 50 60 70 80 90
10
-1
10
0
10
1
10
2
10
3
10
4
K = 1, M = 1
20 dB
10 dB
0 dB
-10 dB

R
e
l
a
t
i
v
e

E
n
e
r
g
y
-
E
f
f
i
c
i
e
n
c
y

(
b
i
t
s
/
J
)
/
(
b
i
t
s
/
J
)

Spectral-Efficiency (bits/s/Hz)
18/44
Erik G. Larsson
Spectral-energy eciency tradeo, cont.

R =
_
_
_
_
1

T
_
K log
2
_
1 +
(M1)P
2
(K1)P
2
+(+K)P+1
_
, for MRC
_
1

T
_
K log
2
_
1 +
(MK)P
2
(+K)P+1
_
, for ZF
Energy eciency: =
R
P
max
P,
, s.t. R = const.
Single-user SIMO system: K = 1, M = 100

max
P,
, s.t. R = const.
19/44
Erik G. Larsson
UL, T = 200, SISO and SU-SIMO
0 10 20 30 40 50 60 70 80 90
10
-1
10
0
10
1
10
2
10
3
10
4
K=1, M=1 (SISO)
20 dB
10 dB
0 dB
-10 dB

R
e
l
a
t
i
v
e

E
n
e
r
g
y
-
E
f
f
i
c
i
e
n
c
y

(
b
i
t
s
/
J
)
/
(
b
i
t
s
/
J
)

K=1, M=100 (SU-SIMO)
20/44
Erik G. Larsson
Spectral-energy eciency tradeo, cont.

R =
_
_
_
_
1

T
_
K log
2
_
1 +
(M1)P
2
(K1)P
2
+(+K)P+1
_
, for MRC
_
1

T
_
K log
2
_
1 +
(MK)P
2
(+K)P+1
_
, for ZF
Energy eciency: =
R
P
max
P,
, s.t. R = const.
Single-user SIMO system: K = 1, M = 100

max
P,
, s.t. R = const.
Multi-user SIMO system: M = 100, K 1 adapted

max
P,,K
, s.t. R = const.
21/44
Erik G. Larsson
UL, T = 200, SISO, SU-SIMO & MU-SIMO
0 10 20 30 40 50 60 70 80 90
10
-1
10
0
10
1
10
2
10
3
10
4
(MU-SIMO)
M = 100
K adaptive
ZF processing
K=1, M=1 (SISO)
20 dB
10 dB
0 dB
-10 dB
-20 dB

R
e
l
a
t
i
v
e

E
n
e
r
g
y
-
E
f
f
i
c
i
e
n
c
y

(
b
i
t
s
/
J
)
/
(
b
i
t
s
/
J
)

K=1, M=100 (SU-SIMO)
MRC processing
22/44
Erik G. Larsson
Analysis Methodology - MRC & Perfect CSI
y =
PGGGxxx + nnn
..
noise, i.i.d. CN(0,1)
z
mrc
GGG
H
yyy = GGG
H
_
PGGGxxx +nnn
_
=
PG
H
Gx +G
H
n
z
k,mrc
=
Pggg
k
2
x
k
. .
desired signal
+
i=k
ggg
H
k
ggg
i
x
i
. .
interference
+ggg
H
k
nnn
..
noise
Achievable ergodic rate

R
k,mrc
=
E
_
log
2
_
1 +
Pggg
k
2
P
K
i=k
| g
i
|
2
+ 1
__
log
2
_
_
1 +
_
E
_
P
K
i=k
| g
i
|
2
+ 1
Pggg
k
2
__
1
_
_
= log
2
_
1 +
P (M 1)
P(K 1) + 1
_
Observe that: g
i

ggg
H
k
ggg
i
ggg
k
CN (0, 1), indep. of g

k
Use E[log(1 + 1/x)] log(1 + 1/E[x])
Here: E
1
g
k
=
1
M1
, since
E
tr
WWW
1
=
m
nm
, if
WWW W
m
(n, III
n
).
23/44
Erik G. Larsson
Analysis Methodology - MRC & Estimated CSI
CSI from pilots:

GGG = GGG +EEE, where
i
CN
_
0,
1
P
p
+1
III
M
_
, ggg
i
CN
_
0,
P
p
P
p
+1
III
M
_
,
P
p
= P, =#pilots
MRC
z
mrc
=

GGG
H
yyy =

G
H
(
PGx +n) =

GGG
H
_
P

GGGxxx
PEEExxx +nnn
_
z
k,mrc
=
P ggg
k
2
x
k
+
P
K
i=k
ggg
H
k
ggg
i
x
i
P
K
i=1
ggg
H
k

i
x
i
+ ggg
H
k
nnn
Achievable ergodic rate per terminal

R
k,mrc
=
_
1

T
_
E
_
log
2
_
1 +
P ggg
k
2
P
K
i=k
|
g
i
|
2
+ P
K
i=1
1
P+1
+ 1
__
_
1

T
_
log
2
_
_
1 +
_
E
_
P
K
i=k
|
g
i
|
2
+ P
K
i=1
1
P+1
+ 1
P ggg
k
2
__
1
_
_
=
_
1

T
_
log
2
_
1 +
P
2
(M 1)
P (P + 1) (K 1) + ( + 1) P + 1
_
g
k
and

g
i

ggg
H
k
ggg
i
ggg
k
, i = k are independent
24/44
Erik G. Larsson
Summary
10 spectral eciency and 1000 radiated energy eciency

possible with M = 100 antennas and MRC
Results for DL are similar (no D.o.F penalty for pilot transmission)
25/44
Erik G. Larsson
Limits Imposed by Intercell Interference
26/44
Erik G. Larsson
Pilot Contamination Problem
(Probably) must rely on UL pilots to get CSI
On UL, may rely partly on blind/joint channel estimation and data

decoding techniques to avoid pilot based estimation
But on DL, need explicit CSI to precode

27/44
Erik G. Larsson
Pilot Contamination Problem, cont.
Consider UL training; one interferer

Y
p
..
M
= G
..
MK
..
K
+N + fff
..
M1
T
..
1
LS estimation, with orthogonal pilots

(
H
= P
p
I, say P
p
= 1)
G = Y
p
H
= G
..
MK
+N
H
. .
i.i.d.
+ fff
..
M1
H
. .
1K
. .
rank1
1
2
M
g1
gk
gK
1
K
k
f
Say g
1
used for MRC data detection: y = Gx +fffz +n
1
M
g
H
1
y =
1
M
_
g
H
1
Gx + g
H
1
fffz + g
H
1
n
_
=
1
M
g
H
1
G
. .
eee
T
1
x
. .
x
1
+
1
M
_
fff
H
_
1
fffz
. .
1
M
(
)
1
fff
H
fffz
. .
(
)
1
z
+
..
O
_
1
M
_
28/44
Erik G. Larsson
Pilot Contamination Problem, cont.
In principle, could project away fff

T
:
Y
p
T = G
T +N
T
and estimate G from Y
p
T . But rank
_
T
_
= 1, so we
sacrice signal space
The eect on the downlink is similar
Can show: p.c. fundamentally limits performance of massive MIMO

if P is xed and M [Mar2010]
Can show: other cells transmitting data is as bad as other cells

sending pilots [NLM2013]
But currently not clear, how bad the p.c. eect is if other types of
CSI estimation is used
29/44
Erik G. Larsson
Summary
Interference from other cells can signicantly impair CSI estimation

and the ensuing UL detection & DL precoding: pilot
contamination
Possible (partial) countermeasures (on UL)
Blind (partial pilot-free) CSI estimation (NL2011)
identify subspace associated with dierent cells exploiting power

control & channel hardening [MVC2013]
clever allocation of pilots in dierent cells [YGFL2013,FAM2013]
pilot cont. precoding [AM2012]

30/44
Erik G. Larsson
Excess Degrees of Freedom in Massive MIMO:
A Unique Opportunity for Hardware-Friendly Signal
Shaping
31/44
Erik G. Larsson
Excess Degrees of Freedom
There are M K unused degrees of freedom.
In the downlink, these excess DoF may be used to shape the

transmitted signals in a hardware-friendly way
Per-antenna constant envelope or low-PAR multiuser precoding
Robust to PA non-linearity/less PA backo
Exploit nullspace of H
T
:
nullspace dimension = M K!
y = H
T
x +w = H
T
(x +z) +w
32/44
Erik G. Larsson
Beamforming versus Low-PAR Signal Shaping
33/44
Erik G. Larsson
Constant-env. MU-MIMO, Flat channel [ML2013]
0 50 100 150 200 250 300
10
8
6
4
2
0
2
4
6
8
Number of Base Station Antennas (M)
M
i
n
.

r
e
q
d
.

P
T
/
2

t
o

a
c
h
i
e
v
e

a

p
e
r
u
s
e
r

r
a
t
e

o
f

2

b
p
c
u

K = 10, Proposed CE Precoder (CE)
K = 10, GBC Sum Cap. Upp. Bou. (APC)
1.7 dB
34/44
Erik G. Larsson
Constant-env. MU-MIMO, Freq. sel. channel [ML2013a]
0 10 20 30 40 50 60
4
3
2
1
0
1
2
3

M
i
n
.

r
e
q
d
.

P
T
/
2

(
d
B
)

t
o

a
c
h
i
e
v
e

2

b
p
c
u
/
u
s
e
r

L=1,2,4,8. ZF upper bound (TAPC)
L = 1,2,4,8. Coop. lower bound (TAPC)
L = 1, CE
L = 2, CE (T >> )
L = 4, CE (T >> )
L = 8, CE (T >> )
35/44
Erik G. Larsson
CE-MU-MIMO PrecodingAnalysis Methodology
E
k
: energy per information symbol u
k
(per user)
P: total transmit power
Received signals:
y
k
=
_
P
M
M
i=1
h
k,i
e
j
i
+n
k
=
P
_
E
k
u
k
+
P
_
M
i=1
h
k,i
e
j
i
_
E
k
u
k
_
. .
k
+n
k
With Gaussian symbols per user:

I(y
k
; u
k
) = h(u
k
) h(u
k
| y
k
) = h(u
k
) h
_
u
k
y
k
E
k
y
k
_
h(u
k
) h
_
u
k
y
k
E
k
_
h(u
k
) h
_

k
E
k
+
n
k
E
k
_
= log
2
(e) h
_

k
E
k
+
n
k
E
k
_
log
2
(e) log
2
_
e var
_

k
E
k
+
n
k
E
k
_
_
log
2
(e) log
2
_
e E
_
E
k
+
n
k
E
k
2
_
_
= log
2
(e) log
2
_
e
_
E[|
k
|
2
]
E
k
+
1
PE
k
_
_
Downlink signal design:

min
i
[,)
K
k=1
|
k
|
2
min
i
[,)
K
k=1
M
i=1
h
k,i
e
j
i
_
E
k
u
k
2
36/44
Erik G. Larsson
PAR-Aware MU-MIMO OFDM Downlink [SL2013]
Precoder design problem:

(PMP)
_
_
minimize max
_
a
1
, . . . , a
N
_
subject to y
w
= H
w
s
w
+ noise, w T
y
w
= only noise, w T
c
.
H
w
: time-frequency channel
T : used subcarriers; T
c
: null subcarriers
Convex optimization problem, fast algorithm; relaxation

37/44
Erik G. Larsson
PAR-aware MU-MIMO OFDM DL (99% percentiles)
38/44
Erik G. Larsson
PAR-aware MU-MIMO OFDM DL (99% percentiles)
39/44
Erik G. Larsson
Summary
Low PAR may be desirable to enable high overall energy eciency &
low cost
Large number of antennas oers entirely new and unique

opportunities for low PAR signal design
40/44
Erik G. Larsson
Massive MIMOa Goldmine of Research Problems
Base station receiver processing
Partially distributed processing (MRC/ZF/MMSE hybrids)
Advanced channel estimation (exploit excess DoF)
Synchronization at low SNR
Base station transmitter processing
Partially distributed processing
Low-complexity low-PAPR precoders
OFDM or single-carrier?
Eects of hardware imperfections: Will the law of large numbers

deliver?
phase noise and clock distribution,
I/Q imbalance, A/D resolution, PA linearity
Pilot contamination (both UL & DL) may be mitigated by
Blind channel estimation algorithms [NL2012]
Coordination and planning of pilot reuse [YGFL2013]
TDD operation using UL channels for DL precoding requires

reciprocity calibration
Cost? (BW & power), algorithms?, dedicated hardware?
Other system issues
Time and frequency synchronization at low SNR
Idle-terminal paging (DL) under non-CSI@TX

41/44
Erik G. Larsson
Eq.-Free SC vs. OFDM @ 1 bps/Hz/t [PML2012]
50 100 150 200 250 300 350 400
18
16
14
12
10
8
6
4
2
0
No of Base Station Antennas (M)
P
o
w
e
r

(
n
o
r
m
a
l
i
z
e
d
)

[
d
B
]

Time reversal MRT precoding
Coop. SumCapacity Bound
OFDM Bound T
cp
=T
u
/4
K=10
42/44
Erik G. Larsson
Literature
RPL+2013 F. Rusek, D. Persson, B. K. Lau, E. G. Larsson, T. L.
Marzetta, O. Edfors, and F. Tufvesson, Scaling up MIMO:
Opportunities and Challenges with Large Arrays, IEEE SP
Magazine, Jan. 2013
NLM2013 H.Q. Ngo, E.G. Larsson and T. Marzetta, Energy and
Spectral Eciency of Very Large Multiuser MIMO
Systems, IEEE Trans. Comm. 2013, arXiv:1112.3810.
ML2013 S.K. Mohammed and E.G. Larsson, Per-antenna Constant
Envelope Precoding for Large Multi-User MIMO Systems,
IEEE Trans. Comm. 2013, arXiv:1201.1634v1.
ML2013a S.K. Mohammed and E.G. Larsson, Constant-Envelope
Multi-User Precoding for Frequency-Selective Massive
MIMO Systems, arXiv:1305.1525.
SL2013 C. Studer and E.G. Larsson, PAR-Aware Large-Scale
Multi-User MIMO-OFDM Downlink, IEEE JSAC, Feb.
2013.
HBD2013 J. Hoydis, S. ten Brink, M. Debbah, Massive MIMO in the
UL/DL of cellular networks: How many antennas do we
need, IEEE JSAC, Feb. 2013.
GERT2011 X. Gao, O. Edfors, F. Rusek, and F. Tufvesson, Linear
pre-coding performance in measured very-large MIMO
channels, IEEE VTC 2011
GTER2012 X. Gao, F. Tufvesson, O. Edfors, and F. Rusek, Measured
propagation characteristics for very-large MIMO at 2.6
GHz, in Proc. Asilomar, 2012 VTC 2011
YGFL2013 H. Yin, D. Gesbert, M. Filippou, and Y. Liu, A
Coordinated Approach to Channel Estimation in Large-scale
Multiple-antenna Systems, IEEE JSAC, Feb. 2013.
AM2012 A. Ashikhmin and T. Marzetta, Pilot Contamination
Precoding in Multi-Cell Large Scale Antenna Systems, in
Proc. ISIT 2012.
NL2012 H. Q. Ngo and E. G. Larsson, EVD-Based Channel
Estimations for Multicell Multiuser MIMO with Very Large
Antenna Arrays, ICASSP 2012.
GJ2011 B. Gopalakrishnan and N. Jindal, An Analysis of Pilot
Contamination on Multi-User MIMO Cellular Systems with
Many Antennas, IEEE SPAWC 2011.
NML2013 H. Q. Ngo, T. Marzetta and E. G. Larsson, The multicell
multiuser MIMO uplink with very large antenna arrays and
a nite-dimensional channel, IEEE TCOM, 2013.
PML2012 A. Pitarokoilis, S. K. Mohammed and E. G. Larsson, On
the optimality of single-carrier transmission in large scale
antenna systems, IEEE Wireless Communication Letters,
2012.
PML2012b A. Pitarokoilis, S. K. Mohammed and E. G. Larsson, Eect
of Oscillator Phase Noise on Uplink Performance of Large
MU-MIMO Systems, in Proc. Allerton, 2012.
Mar2010 T. L. Marzetta, Noncooperative MU-MIMO with unlimited
numbers of base station antennas, IEEE Trans. WC 2010.
JAMV2011 J. Jose, A. Ashikhmin, T. L. Marzetta, and S. Vishwanath,
Pilot contamination and precoding in multi-cell TDD
systems, IEEE Trans. WC 2011.
ZYP2013 J. Zhang, X. Yuan, and L. Ping, Hermitian Precoding for
Distributed MIMO Systems with Individual Channel State
Information, IEEE JSAC 2013.
SY...2012 C. Shepard et al., Argos: practical many-antenna base
stations, in Proc. MobiCom 2012.
MVC2013 R. R. M uller, M. Vehkapera, and L. Cottatellucci, Blind
Pilot Decontamination, in Proc. WSA 2013.
FAM2013 F. Fernandes, A. Ashikhmin, and T. L. Marzetta, Inter-Cell
Interference in Noncooperative TDD Large Scale Antenna
Systems, IEEE JSAC, Feb. 2013.
LTEM2013 E. G. Larsson, F. Tufvesson, O. Edfors and T. L. Marzetta,
Massive MIMO for Next Generation Wireless Systems,
arXiv:1304.6690.
43/44
Erik G. Larsson
Thank You
Visit the massive MIMO website
www.commsys.isy.liu.se/egl/vlm/vlm.html
44/44
Erik G. Larsson

Massive MIMO: Fundamentals, Opportunities and Challenges: MM YS

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Massive MIMO: Fundamentals, Opportunities and Challenges: MM YS

Загружено:

Авторское право:

Доступные форматы

MM

Massive multiuser MIMO (MISO):

M K 1 (think 100 10 or 500 50)

coherent, but simple, processing

Potential to dramatically improve rate & reliability

Potential to drastically scale down TX power

Not only theory, at least one known testbed (64 10)

Reduce bulky items (coax)

Each antenna unit simple (low accuracy)

Resilience against individual failures (hotswapping)

Potential economy of scale in manufacturing

acquire CSI from uplink pilots and/or blindly from data

M K linear processing (MRC, ZF, MMSE) nearly optimal

use CSI obtained on the uplink

make necessary adjustments based on reciprocity calibration

apply multiuser MIMO precoding

intracell interference will appear as noise

1 bps/Hz/terminal; K bps/Hz/terminal total

can cancel out intra cell interference

computationally more demanding

Multiplexing gain K, Array power gain M (ideally)

Rely on the law of large numbers

average out fast fading and thermal noise

no FD scheduling, full BW to all terminals, simple MAC

almost no air-interface latency

In the 1 bps/Hz/t regime, many impairments drown in thermal noise

M K unused degrees of freedom (e.g. 500 50 = 450)

Multiplexing gain can only materialize if channel responses are

Array gain relies on coherency getting CSI is the main thing

HW power consumption (RF) must scale fast enough with P

Scaling down P thermal noise eventually limits performance.

P large enough, interference limits performance

Favorable propagation (f.p.) if

Our partners at Lund Univ., Sweden have conducted unique

Indoor 128-ant. (4x16 dual-pol.) array. 3 users indoor, 3 outdoor.

2.6 GHz CF, 50 MHz BW, 100 snapshots (10m).

Normalized to retain only small-scale fading.

Assumptions on f.p. have substantial support in measurements

I.i.d. model for small-scaled fading appears reasonable

More details and models in F. Tufvessons tutorial next

Power (SNR) per terminal: P

i.i.d. Rayleigh fading

Coherence interval: T symbols (e.g. 2 100kHz 1ms = 200)

K mutually orthogonal pilots of length (K T)

MMSE channel estimation

Sum-spectral eciency bounds [NLM2013]:

Sum-spectral eciency bounds [NLM2013]:

Single-user SIMO system: K = 1, M = 100

Sum-spectral eciency bounds [NLM2013]:

Single-user SIMO system: K = 1, M = 100

Multi-user SIMO system: M = 100, K 1 adapted

Achievable ergodic rate

CN (0, 1), indep. of g

Use E[log(1 + 1/x)] log(1 + 1/E[x])

CSI from pilots:

Achievable ergodic rate per terminal

10 spectral eciency and 1000 radiated energy eciency

(Probably) must rely on UL pilots to get CSI

On UL, may rely partly on blind/joint channel estimation and data

But on DL, need explicit CSI to precode

Consider UL training; one interferer

LS estimation, with orthogonal pilots

In principle, could project away fff